Stacked CNN-based multichannel attention networks for Alzheimer disease detection

Area of Science:

• Neurology
• Artificial Intelligence
• Medical Imaging

Background:

• Alzheimer's Disease (AD) is a progressive neurological disorder characterized by cognitive decline.
• Current deep learning models for AD classification face challenges with accuracy and efficiency due to feature ineffectiveness.
• Automatic diagnosis systems for AD using medical images are crucial for early intervention.

Purpose of the Study:

• To propose a lightweight Stacked Convolutional Neural Network with a Channel Attention Network (SCCAN) for improved Alzheimer's Disease classification.
• To enhance feature extraction and selection for more effective AD recognition from MRI data.
• To develop a deep learning model suitable for training on smaller datasets and addressing class imbalance.

Main Methods:

• A novel SCCAN model integrating 5 CNN modules for hierarchical feature extraction.
• A channel attention module to select influential features based on channel dimension.
• Training and validation on diverse MRI datasets including ADNI1, Kaggle, and OASIS-1, addressing class imbalance.

Main Results:

• The SCCAN model demonstrated high classification accuracy: 99.58% (ADNI1), 99.22% (Kaggle), and 99.70% (OASIS-1).
• The model exhibits reduced parameters, making it efficient for training.
• Performance surpassed state-of-the-art models in Alzheimer's Disease classification.

Conclusions:

• The proposed SCCAN model offers a significant advancement in MRI-based Alzheimer's Disease classification.
• SCCAN provides an effective and accurate solution for automatic AD diagnosis.
• The lightweight design and high accuracy make SCCAN a promising tool for clinical application.